Shaping your AI future: Ten critical questions

Organisations know that they risk being left behind if they fail to embrace AI and automation. But decisions around what, when, how and with whom are difficult to make and far-reaching in impact. In this article we outline ten important considerations ahead of implementing such technology.

In a marketplace dominated by hype, unrealistic promises and obfuscation, finding the best way forward can be challenging. Factor in the bewildering pace of AI innovation and it can be tempting to defer decisions and wait for a point of clarity that is unlikely to ever arrive.

To help make such decisions easier, we put together a panel of experienced subject matter experts, each with practical but different perspectives on AI and automation, to talk through a wide range of scenarios and use-cases.

Richard McKenzie-Small, Mark Baginski and Martin Cropper have, between them, been involved with many automation activities covering multiple aspects of front, mid and back-office change.

We asked them to identify key considerations based on their personal experience; lessons learnt at the ‘coal face’ that would help organisations thinking about investing in such technology make better decisions.

Here are their ten key points your organisation should consider ahead of pushing forward with investment.

  1. Reallyunderstand your customer and/or colleague​

Success, whether measured by uptake/adoption, increased efficiency or quality, or reduced waste or cost, depends on how well a new automated solution is designed.

Experience shows that the best ones are based on a thorough understanding of the specific wants, needs, expectations and behaviours of your customer and / or colleagues. Gathering this insight will also drive choices made when selecting the target use cases.​

  1. Start simple, start small and augment​

All three panel members describe positive experience of projects where the AI/automation use-case could be delivered with existing data or interfaces and wasn’t going to challenge any complex areas of the enterprise rules, regulations or technology.

Such ‘simple’ use cases that will enhance customer or colleague convenience, while delivering rapid and visible benefits, are a great place to start. Once they have been proven and lessons learned, the next tranche of use cases should be easier to get approved.​

  1. Do your research and make a conscious choice of which problem to solve

The panel had different views on where to start with no consensus on which business area and what type of process to initially pursue.

Two suggestions emerged. They may end up with the same answer and are based on potential business impact.

3.1: Find an area where AI & automation will make a difference but where the problems to solve are easily identifiable and relatively straightforward.  The cost and risk of an initial project will be low and the impact will, by definition, be lower too.  This is very often the approach needed in a regulated business.

3.2: Consciously opt to take on a ‘knottier problem’, given that modern AI and automation tools can be applied successfully to problems that would have exceeded the capabilities of previous generations of technology. For example, if there is a complex, dynamic process with multiple data sources, it may be that a tool can be trained (and learn) quickly and prove itself in a pilot before being rolled out more widely. Done right, this approach can demonstrate real pace and impact and, if well-designed and well-managed, be medium-low risk.

  1. Not every automation requires AI​

In the panel’s experience, many smart workflow/automation projects are packaged with AI tools, almost by default.​   Using AI may increase the cost, increase the time to launch and add a level of complexity that isn’t necessarily justified.   Often, a significant proportion of the business benefit can be delivered without AI/ML, perhaps with more advanced subsequent phases to follow once initial lessons have been learned and the business case is gaining momentum.​

  1. Think very carefully before trying to replace people – augment not replace​

There have been several newsworthy ‘fails’ where a complex and risky technology such as GenAI has been let loose on customers before it was ready or properly tested/controlled.   Our advice is to keep a human-in-the-loop for customer-facing automations until they have demonstrated reliability, quality and performance.

This may take the form of a colleague monitoring real-time performance or it may consist of ‘escape routes’ should something go awry.  Internal AI-powered apps should also be carefully tested and proven – Martin shared his experiences of clients trying to replace human interviewers with automatons, rather than supporting human interviewers with better information and prompts. It didn’t go well.​

For now, at least, it is better to help your colleagues be ‘the best that they can be’, using AI to help them perform better, rather than to replace them.

  1. Really understand your data and business rules (and regulations)​

However good the AI/automation tool, there are two things that it won’t work without:   (i) high-quality, available data and (ii) an in-depth understanding of your business rules.

A vital early task in any project will be to map out every data source and any gaps, carefully documenting the decision points, rules and controls that govern your processes.  This will then enable you to assess initial feasibility, support the design, implementation and testing of the tool and assess its quality and performance.

  1. Don’t (just) listen to vendor salespeople​

While this is a generalisation (there are some genuinely good products out there), the sales pitch will typically overhype the pace and benefits, while underplaying the cost, complexity and risk. As experienced implementers of AI- and non-AI powered solutions, our panel had several pieces of advice, for example:​

  • Research beyond vendors – talk to people who have done similar things, even if they are in different sectors, to truly understand the depth and breadth of the challenge
  • Think about how you are going to deliver the solution and fully understand what skills you will need around you​
  • Consider the full array of changes you will need to make; how your customers and colleagues will react​ and how can likely issues be mitigated?
  1. Truly tailored and personalised solutions are real and can add value​

For a long time, we have been warned away from tailored technology solutions with difficult upgrades, compatibility clashes and divergence from best practice all cited as issues. However, a system capable of learning your processes and rules, or digesting all of your training and knowledge articles in a few weeks, will easily be able to adapt when they change.  While testing might be more complex, this might change some of the design principles that underpin transformation projects going forward. ​

Some of the tools available now can generate personalised recommendations (for example, customer profiling and next-best-action in a sales environment).  Martin shared his experience of AI-powered training and coaching tools that learn their ‘client’ quickly and provide genuinely personalised coaching advice.

  1. Take time to consider the people implications of what you are doing​

Mark and Martin pointed out the risks of taking (or trying to take) the human ‘out of the loop’ when there are unclear business rules or important decisions that need to be made.

Careful management of the potentially negative impacts of AI tools (i.e. replacing human colleagues) is also vital for obvious reasons.  Richard highlighted a different aspect of the potential people impacts where the use of automation and knowledge tools in a contact centre is having a positive impact on colleague satisfaction and performance; automated tools have absorbed some of the dull, repetitive work, leaving colleagues to handle more complex (and perhaps more interesting) customer enquiries. ​

  1. Keep looking forward, be curious about what might be coming​ over the horizon

Two of our panellists illustrated this point by highlighting common technologies that five years ago seemed impossible.  Clearly the same is going to be true of AI and Automation in the future.  Martin called-out personalised, augmented skills applications as something we’ll soon see.  Other applications that were ‘pipedreams’ not long ago but are becoming more and more feasible include real-time (voice) translation and AI vision (you can see a great example of this by searching AI-powered shoplifter detection in store).

A non-technician can keep up to date by keeping close to a handful of the hundreds of on-line forums and experts sharing experiences (try the Global Business & AI Executive Forum and linkedin.com/in/pascalbornet for example).

Analysts like Gartner and consultancies such as BCG publish regular market reviews, articles and case studies, which are often enlightening.  Vendors and sales events are also useful, albeit that there is likely to be a bias in how they present features and benefits, with cross-business impact often underplayed.​

Finally, we also suggest you encourage your people to be curious and experiment with the widely available AI platforms (such as CoPilot and ChatGPT) beyond simple queries on a day to day basis. We’d also recommend talking to others about what is working for them.

We hope you find this article helpful. Should you wish to find out more about any of them, or if you need independent guidance in identifying, selecting and implementing new technology, please don’t hesitate to get in touch.